Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A method for building efficient frameworks by directly imitating two-stage features

A staged, direct technology, applied in the field of building efficient frameworks by directly imitating two-stage features, can solve problems such as large accuracy gaps, achieve high efficiency and high precision, high efficiency, and overcome feature asymmetry.

Active Publication Date: 2021-03-16
FOSHAN NANHAI GUANGDONG TECH UNIV CNC EQUIP COOP INNOVATION INST
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Although recent work attempts to improve one-stage detectors by imitating the structural design of two-stage detectors, the accuracy gap is still large

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A method for building efficient frameworks by directly imitating two-stage features
  • A method for building efficient frameworks by directly imitating two-stage features
  • A method for building efficient frameworks by directly imitating two-stage features

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. All other embodiments obtained by ordinary persons in the art without creative efforts belong to the protection scope of the present invention.

[0030] This implementation implements a novel and efficient framework for training one-stage detectors by directly imitating two-stage features, using a shared feature pyramid backbone network to extract high-quality features, and using the results obtained by learning these features with a two-stage detector to guide the one-stage detector. The training process specifically includes the following implementation steps:

[0031] Step 1. Construct the model's feature pyramid network backbone network with resnet101 and FPN network...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The present invention provides a method for building an efficient framework by directly imitating two-stage features, including: S1, constructing a modeled feature pyramid network backbone network with resnet101 and FPN network; S2, after extracting features in FPN, using the Refinement module to filter out negative effects , adjust the position and size of the predefined anchor box; S3, the branch of the two-stage detection head, detect the sparse set of anchor boxes adjusted by the Refinement module, and divide T-head into two branches for classification and regression; S4, the one-stage detection head branch, design it as a lightweight network; S5, define the training loss function, improve the accuracy of the first-stage detector, make it easier for the first-stage detector to obtain useful information, and make it easier to obtain useful information without increasing the calculation cost Under this condition, the high precision of the two-stage detection head and the high efficiency of the one-stage detection head can be obtained.

Description

technical field [0001] The invention relates to the field of deep learning computer vision, in particular to a method for building an efficient framework by directly imitating two-stage features. Background technique [0002] Existing object detection methods can be divided into one-stage methods and two-stage methods. One-stage detectors are more efficient due to their simple architecture, while two-stage detectors lead in terms of accuracy due to their structures that generate more accurate candidate boxes. Although recent works try to improve one-stage detectors by imitating the structural design of two-stage detectors, their accuracy gap is still large. We propose a novel and efficient framework for training one-stage detectors by directly imitating two-stage features, aiming to bridge the accuracy gap between one-stage and two-stage detectors. Different from traditional analog methods, this method has a shared backbone for one-stage and two-stage detectors, which is t...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/32G06N3/04
CPCG06V10/25G06N3/045
Inventor 李泽辉杨淑爱李俊宇黄坤山
Owner FOSHAN NANHAI GUANGDONG TECH UNIV CNC EQUIP COOP INNOVATION INST
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products